2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1990 The Regents of the University of California.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. Neither the name of the University nor the names of its contributors
15 * may be used to endorse or promote products derived from this software
16 * without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include "opt_capsicum.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/capsicum.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
48 #include <sys/sysproto.h>
53 #include <vm/vm_kern.h> /* for kernel_map */
54 #include <vm/vm_extern.h>
56 #include <machine/frame.h>
57 #include <machine/md_var.h>
58 #include <machine/pcb.h>
59 #include <machine/specialreg.h>
60 #include <machine/sysarch.h>
61 #include <machine/tss.h>
62 #include <machine/vmparam.h>
64 #include <security/audit/audit.h>
66 static void user_ldt_deref(struct proc_ldt *pldt);
67 static void user_ldt_derefl(struct proc_ldt *pldt);
71 int max_ldt_segment = 512;
72 SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN,
74 "Maximum number of allowed LDT segments in the single address space");
77 max_ldt_segment_init(void *arg __unused)
80 if (max_ldt_segment <= 0)
82 if (max_ldt_segment > MAX_LD)
83 max_ldt_segment = MAX_LD;
85 SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
87 #ifndef _SYS_SYSPROTO_H_
95 sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
97 struct i386_ldt_args *largs, la;
98 struct user_segment_descriptor *lp;
102 * XXXKIB check that the BSM generation code knows to encode
105 AUDIT_ARG_CMD(uap->op);
106 if (uap_space == UIO_USERSPACE) {
107 error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
112 largs = (struct i386_ldt_args *)uap->parms;
116 error = amd64_get_ldt(td, largs);
119 if (largs->descs != NULL && largs->num > max_ldt_segment)
121 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
122 if (largs->descs != NULL) {
123 lp = malloc(largs->num * sizeof(struct
124 user_segment_descriptor), M_TEMP, M_WAITOK);
125 error = copyin(largs->descs, lp, largs->num *
126 sizeof(struct user_segment_descriptor));
128 error = amd64_set_ldt(td, largs, lp);
131 error = amd64_set_ldt(td, largs, NULL);
139 update_gdt_gsbase(struct thread *td, uint32_t base)
141 struct user_segment_descriptor *sd;
145 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
147 sd = PCPU_GET(gs32p);
148 sd->sd_lobase = base & 0xffffff;
149 sd->sd_hibase = (base >> 24) & 0xff;
154 update_gdt_fsbase(struct thread *td, uint32_t base)
156 struct user_segment_descriptor *sd;
160 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
162 sd = PCPU_GET(fs32p);
163 sd->sd_lobase = base & 0xffffff;
164 sd->sd_hibase = (base >> 24) & 0xff;
169 sysarch(struct thread *td, struct sysarch_args *uap)
172 struct pcb *pcb = curthread->td_pcb;
175 struct i386_ioperm_args iargs;
176 struct i386_get_xfpustate i386xfpu;
177 struct amd64_get_xfpustate a64xfpu;
179 #ifdef CAPABILITY_MODE
181 * When adding new operations, add a new case statement here to
182 * explicitly indicate whether or not the operation is safe to
183 * perform in capability mode.
185 if (IN_CAPABILITY_MODE(td)) {
189 case I386_GET_IOPERM:
190 case I386_GET_FSBASE:
191 case I386_SET_FSBASE:
192 case I386_GET_GSBASE:
193 case I386_SET_GSBASE:
194 case I386_GET_XFPUSTATE:
195 case AMD64_GET_FSBASE:
196 case AMD64_SET_FSBASE:
197 case AMD64_GET_GSBASE:
198 case AMD64_SET_GSBASE:
199 case AMD64_GET_XFPUSTATE:
202 case I386_SET_IOPERM:
205 if (KTRPOINT(td, KTR_CAPFAIL))
206 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
213 if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
214 return (sysarch_ldt(td, uap, UIO_USERSPACE));
216 * XXXKIB check that the BSM generation code knows to encode
219 AUDIT_ARG_CMD(uap->op);
221 case I386_GET_IOPERM:
222 case I386_SET_IOPERM:
223 if ((error = copyin(uap->parms, &iargs,
224 sizeof(struct i386_ioperm_args))) != 0)
227 case I386_GET_XFPUSTATE:
228 if ((error = copyin(uap->parms, &i386xfpu,
229 sizeof(struct i386_get_xfpustate))) != 0)
231 a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
232 a64xfpu.len = i386xfpu.len;
234 case AMD64_GET_XFPUSTATE:
235 if ((error = copyin(uap->parms, &a64xfpu,
236 sizeof(struct amd64_get_xfpustate))) != 0)
244 case I386_GET_IOPERM:
245 error = amd64_get_ioperm(td, &iargs);
247 error = copyout(&iargs, uap->parms,
248 sizeof(struct i386_ioperm_args));
250 case I386_SET_IOPERM:
251 error = amd64_set_ioperm(td, &iargs);
253 case I386_GET_FSBASE:
254 update_pcb_bases(pcb);
255 i386base = pcb->pcb_fsbase;
256 error = copyout(&i386base, uap->parms, sizeof(i386base));
258 case I386_SET_FSBASE:
259 error = copyin(uap->parms, &i386base, sizeof(i386base));
261 set_pcb_flags(pcb, PCB_FULL_IRET);
262 pcb->pcb_fsbase = i386base;
263 td->td_frame->tf_fs = _ufssel;
264 update_gdt_fsbase(td, i386base);
267 case I386_GET_GSBASE:
268 update_pcb_bases(pcb);
269 i386base = pcb->pcb_gsbase;
270 error = copyout(&i386base, uap->parms, sizeof(i386base));
272 case I386_SET_GSBASE:
273 error = copyin(uap->parms, &i386base, sizeof(i386base));
275 set_pcb_flags(pcb, PCB_FULL_IRET);
276 pcb->pcb_gsbase = i386base;
277 td->td_frame->tf_gs = _ugssel;
278 update_gdt_gsbase(td, i386base);
281 case AMD64_GET_FSBASE:
282 update_pcb_bases(pcb);
283 error = copyout(&pcb->pcb_fsbase, uap->parms,
284 sizeof(pcb->pcb_fsbase));
287 case AMD64_SET_FSBASE:
288 error = copyin(uap->parms, &a64base, sizeof(a64base));
290 if (a64base < VM_MAXUSER_ADDRESS) {
291 set_pcb_flags(pcb, PCB_FULL_IRET);
292 pcb->pcb_fsbase = a64base;
293 td->td_frame->tf_fs = _ufssel;
299 case AMD64_GET_GSBASE:
300 update_pcb_bases(pcb);
301 error = copyout(&pcb->pcb_gsbase, uap->parms,
302 sizeof(pcb->pcb_gsbase));
305 case AMD64_SET_GSBASE:
306 error = copyin(uap->parms, &a64base, sizeof(a64base));
308 if (a64base < VM_MAXUSER_ADDRESS) {
309 set_pcb_flags(pcb, PCB_FULL_IRET);
310 pcb->pcb_gsbase = a64base;
311 td->td_frame->tf_gs = _ugssel;
317 case I386_GET_XFPUSTATE:
318 case AMD64_GET_XFPUSTATE:
319 if (a64xfpu.len > cpu_max_ext_state_size -
320 sizeof(struct savefpu))
323 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
324 a64xfpu.addr, a64xfpu.len);
335 amd64_set_ioperm(td, uap)
337 struct i386_ioperm_args *uap;
340 struct amd64tss *tssp;
341 struct system_segment_descriptor *tss_sd;
346 if ((error = priv_check(td, PRIV_IO)) != 0)
348 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
350 if (uap->start > uap->start + uap->length ||
351 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
356 * While this is restricted to root, we should probably figure out
357 * whether any other driver is using this i/o address, as so not to
358 * cause confusion. This probably requires a global 'usage registry'.
361 if (pcb->pcb_tssp == NULL) {
362 tssp = (struct amd64tss *)kmem_malloc(kernel_arena,
363 ctob(IOPAGES + 1), M_WAITOK);
364 pmap_pti_add_kva((vm_offset_t)tssp, (vm_offset_t)tssp +
365 ctob(IOPAGES + 1), false);
366 iomap = (char *)&tssp[1];
367 memset(iomap, 0xff, IOPERM_BITMAP_SIZE);
369 /* Takes care of tss_rsp0. */
370 memcpy(tssp, &common_tss[PCPU_GET(cpuid)],
371 sizeof(struct amd64tss));
372 tssp->tss_iobase = sizeof(*tssp);
373 pcb->pcb_tssp = tssp;
374 tss_sd = PCPU_GET(tss);
375 tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
376 tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
377 tss_sd->sd_type = SDT_SYSTSS;
378 ltr(GSEL(GPROC0_SEL, SEL_KPL));
379 PCPU_SET(tssp, tssp);
382 iomap = (char *)&pcb->pcb_tssp[1];
383 for (i = uap->start; i < uap->start + uap->length; i++) {
385 iomap[i >> 3] &= ~(1 << (i & 7));
387 iomap[i >> 3] |= (1 << (i & 7));
393 amd64_get_ioperm(td, uap)
395 struct i386_ioperm_args *uap;
400 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
402 if (td->td_pcb->pcb_tssp == NULL) {
407 iomap = (char *)&td->td_pcb->pcb_tssp[1];
410 state = (iomap[i >> 3] >> (i & 7)) & 1;
411 uap->enable = !state;
414 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
415 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
425 * Update the GDT entry pointing to the LDT to point to the LDT of the
429 set_user_ldt(struct mdproc *mdp)
432 *PCPU_GET(ldt) = mdp->md_ldt_sd;
433 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
437 set_user_ldt_rv(struct vmspace *vmsp)
442 if (vmsp != td->td_proc->p_vmspace)
445 set_user_ldt(&td->td_proc->p_md);
449 user_ldt_alloc(struct proc *p, int force)
451 struct proc_ldt *pldt, *new_ldt;
453 struct soft_segment_descriptor sldt;
457 mtx_assert(&dt_lock, MA_OWNED);
459 if (!force && mdp->md_ldt != NULL)
460 return (mdp->md_ldt);
461 mtx_unlock(&dt_lock);
462 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
463 sz = max_ldt_segment * sizeof(struct user_segment_descriptor);
464 sva = kmem_malloc(kernel_arena, sz, M_WAITOK | M_ZERO);
465 new_ldt->ldt_base = (caddr_t)sva;
466 pmap_pti_add_kva(sva, sva + sz, false);
467 new_ldt->ldt_refcnt = 1;
469 sldt.ssd_limit = sz - 1;
470 sldt.ssd_type = SDT_SYSLDT;
471 sldt.ssd_dpl = SEL_KPL;
478 if (pldt != NULL && !force) {
479 pmap_pti_remove_kva(sva, sva + sz);
480 kmem_free(kernel_arena, sva, sz);
481 free(new_ldt, M_SUBPROC);
486 bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
487 sizeof(struct user_segment_descriptor));
488 user_ldt_derefl(pldt);
491 ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
492 atomic_thread_fence_rel();
493 mdp->md_ldt = new_ldt;
495 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv, NULL,
498 return (mdp->md_ldt);
502 user_ldt_free(struct thread *td)
504 struct proc *p = td->td_proc;
505 struct mdproc *mdp = &p->p_md;
506 struct proc_ldt *pldt;
508 mtx_assert(&dt_lock, MA_OWNED);
509 if ((pldt = mdp->md_ldt) == NULL) {
510 mtx_unlock(&dt_lock);
516 atomic_thread_fence_rel();
517 bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
519 lldt(GSEL(GNULL_SEL, SEL_KPL));
521 user_ldt_deref(pldt);
525 user_ldt_derefl(struct proc_ldt *pldt)
530 if (--pldt->ldt_refcnt == 0) {
531 sva = (vm_offset_t)pldt->ldt_base;
532 sz = max_ldt_segment * sizeof(struct user_segment_descriptor);
533 pmap_pti_remove_kva(sva, sva + sz);
534 kmem_free(kernel_arena, sva, sz);
535 free(pldt, M_SUBPROC);
540 user_ldt_deref(struct proc_ldt *pldt)
543 mtx_assert(&dt_lock, MA_OWNED);
544 user_ldt_derefl(pldt);
545 mtx_unlock(&dt_lock);
549 * Note for the authors of compat layers (linux, etc): copyout() in
550 * the function below is not a problem since it presents data in
551 * arch-specific format (i.e. i386-specific in this case), not in
552 * the OS-specific one.
555 amd64_get_ldt(struct thread *td, struct i386_ldt_args *uap)
557 struct proc_ldt *pldt;
558 struct user_segment_descriptor *lp;
564 printf("amd64_get_ldt: start=%u num=%u descs=%p\n",
565 uap->start, uap->num, (void *)uap->descs);
568 pldt = td->td_proc->p_md.md_ldt;
569 if (pldt == NULL || uap->start >= max_ldt_segment || uap->num == 0) {
570 td->td_retval[0] = 0;
573 num = min(uap->num, max_ldt_segment - uap->start);
574 lp = &((struct user_segment_descriptor *)(pldt->ldt_base))[uap->start];
575 data = malloc(num * sizeof(struct user_segment_descriptor), M_TEMP,
578 for (i = 0; i < num; i++)
579 data[i] = ((volatile uint64_t *)lp)[i];
580 mtx_unlock(&dt_lock);
581 error = copyout(data, uap->descs, num *
582 sizeof(struct user_segment_descriptor));
585 td->td_retval[0] = num;
590 amd64_set_ldt(struct thread *td, struct i386_ldt_args *uap,
591 struct user_segment_descriptor *descs)
594 struct proc_ldt *pldt;
595 struct user_segment_descriptor *dp;
601 printf("amd64_set_ldt: start=%u num=%u descs=%p\n",
602 uap->start, uap->num, (void *)uap->descs);
604 mdp = &td->td_proc->p_md;
607 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
610 /* Free descriptors */
611 if (uap->start == 0 && uap->num == 0)
612 uap->num = max_ldt_segment;
615 if ((pldt = mdp->md_ldt) == NULL ||
616 uap->start >= max_ldt_segment)
618 largest_ld = uap->start + uap->num;
619 if (largest_ld > max_ldt_segment)
620 largest_ld = max_ldt_segment;
621 if (largest_ld < uap->start)
624 for (i = uap->start; i < largest_ld; i++)
625 ((volatile uint64_t *)(pldt->ldt_base))[i] = 0;
626 mtx_unlock(&dt_lock);
630 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
631 /* verify range of descriptors to modify */
632 largest_ld = uap->start + uap->num;
633 if (uap->start >= max_ldt_segment ||
634 largest_ld > max_ldt_segment ||
635 largest_ld < uap->start)
639 /* Check descriptors for access violations */
640 for (i = 0; i < uap->num; i++) {
643 switch (dp->sd_type) {
644 case SDT_SYSNULL: /* system null */
664 /* memory segment types */
665 case SDT_MEMEC: /* memory execute only conforming */
666 case SDT_MEMEAC: /* memory execute only accessed conforming */
667 case SDT_MEMERC: /* memory execute read conforming */
668 case SDT_MEMERAC: /* memory execute read accessed conforming */
669 /* Must be "present" if executable and conforming. */
673 case SDT_MEMRO: /* memory read only */
674 case SDT_MEMROA: /* memory read only accessed */
675 case SDT_MEMRW: /* memory read write */
676 case SDT_MEMRWA: /* memory read write accessed */
677 case SDT_MEMROD: /* memory read only expand dwn limit */
678 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
679 case SDT_MEMRWD: /* memory read write expand dwn limit */
680 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
681 case SDT_MEME: /* memory execute only */
682 case SDT_MEMEA: /* memory execute only accessed */
683 case SDT_MEMER: /* memory execute read */
684 case SDT_MEMERA: /* memory execute read accessed */
690 /* Only user (ring-3) descriptors may be present. */
691 if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
695 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
696 /* Allocate a free slot */
698 pldt = user_ldt_alloc(p, 0);
700 mtx_unlock(&dt_lock);
705 * start scanning a bit up to leave room for NVidia and
706 * Wine, which still user the "Blat" method of allocation.
709 dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
710 for (; i < max_ldt_segment; ++i, ++dp) {
711 if (dp->sd_type == SDT_SYSNULL)
714 if (i >= max_ldt_segment) {
715 mtx_unlock(&dt_lock);
719 error = amd64_set_ldt_data(td, i, 1, descs);
720 mtx_unlock(&dt_lock);
722 largest_ld = uap->start + uap->num;
723 if (largest_ld > max_ldt_segment)
726 if (user_ldt_alloc(p, 0) != NULL) {
727 error = amd64_set_ldt_data(td, uap->start, uap->num,
730 mtx_unlock(&dt_lock);
733 td->td_retval[0] = uap->start;
738 amd64_set_ldt_data(struct thread *td, int start, int num,
739 struct user_segment_descriptor *descs)
742 struct proc_ldt *pldt;
743 volatile uint64_t *dst, *src;
746 mtx_assert(&dt_lock, MA_OWNED);
748 mdp = &td->td_proc->p_md;
750 dst = (volatile uint64_t *)(pldt->ldt_base);
751 src = (volatile uint64_t *)descs;
752 for (i = 0; i < num; i++)
753 dst[start + i] = src[i];